In recent years, Americans have increasingly seen how AI can monitor their movements, associations, and speech. Government agencies have increasingly used AI tools to monitor the social media accounts of protesters, students, and journalists. AI has enabled law enforcement to sift through online activity, using social media posts, likes, comments, follows, and tags as raw material for inferring their political values.
When the government has access to tools that track sentiment, that capacity carries risks, including targeted surveillance of viewpoints a government authority finds politically unpalatable. Rather than following a suspect’s trail through public posts, population-scale sentiment scoring works like a dragnet: scanning everyone and flagging viewpoints, not actions, for law enforcement attention.
A particularly concerning risk of this dragnet is the chilling effect on freedom of expression. To mitigate this threat, Congress should prohibit any law enforcement agency receiving federal dollars from using or procuring AI tools that conduct sentiment analysis across whole populations’ online speech.
For many people, their social media accounts represent a one-stop shop for following current events, voicing their ideas in public, and interacting with friends, colleagues, and others that share their worldview. Since the rise of social media, law enforcement has used social media data as evidence, but investigations have historically been constrained by human capacity and the limited accuracy and analytical capacity of early-generation machine learning systems.
With advances in natural language processing, computer vision, and deep learning, automated analysis systems built for law enforcement have improved at assigning meaning and intent to posts, e.g., classifying them as positive, negative, or threatening. The threat such capability poses lies in its scope, scale, and speed, not just in the kind of analysis it can conduct on individual posts.
When people perceive that they are being surveilled digitally, they may be less inclined to openly publish critical opinions for fear of retribution. In the long run, this dynamic may be detrimental to the civil discourse needed for a functioning democracy. Freedom House, a nonprofit that publishes surveys assessing a country’s democratic health, lowered the U.S. internet-freedom score in 2025, citing, among other factors, federal government plans to deploy AI-enabled social media surveillance.
AI-enabled social media sentiment surveillance is not hypothetical: it is happening now and targets viewpoints across the political spectrum.
In March 2021, the U.S. Postal Service’s Internet Covert Operations Program used AI-enabled tools to track narratives in social media posts from right-wing-leaning Parler and Telegram accounts about protests against 5G and lockdown mandates. In 2025, Immigration and Customs Enforcement issued a call for contracts to use AI for “monitoring and analysis of behavioral and social media sentiment (i.e., positive, neutral, and negative),” and the State Department announced it would use AI to scrutinize student-visa holders’ social media.
These powerful tools are not limited to federal law enforcement. In D.C., local police used the commercial platform Dataminr to monitor protests, receiving 160,000 alerts between June 2020 and May 2022 on demonstrations against issues ranging from COVID restrictions to police brutality.
The Supreme Court has long recognized that government knowledge of people’s associations can be chilling to protected First Amendment activities. However, in practice, it has set a high bar for proving any resulting material chilling effects. Additionally, individual federal agencies often lack robust rules and guardrails for monitoring online expression. To adequately protect freedom of expression, Congress must supply what the judiciary and executive have not.
In particular, legislation must impose firm controls on population-scale AI sentiment scoring. Federal agencies and state and local departments receiving federal funding must be prohibited from using or procuring an AI tool to perform sentiment analysis across a population, demographic, or geographic area. Importantly, “sentiment analysis” must not be defined by how companies market their tools but by what their tools can actually do.
This should include any AI tool that:
A prohibition on this type of AI-enabled analysis is especially necessary because any chilling effect on expression can arise from people’s perceived existence of the capability to autonomously surveil. As such, merely instituting an auditing or disclosure requirement for batch sentiment analysis would likely only provide transparency to the application of such a capability, without remediating that capability’s expression-deterring effects. That is why the capability itself, not its misuse, must be restricted.
Preserving our country’s democratic principles–including freedom of expression–demands red lines in an era in which unrestricted AI use in law enforcement has the potential to create a panopticon of surveillance. Agencies should certainly be allowed to use social media sources to investigate and mitigate individualized threats based on reasonable suspicion, but AI-powered tools that sweep up Americans’ expressed opinions, values, and sentiments are fundamentally at odds with freedom of expression.
AI R&D automation has the potential to transform AI progress, ushering in systems far more capable than today’s at a greater pace. How large that impact will be remains uncertain. But recent reports suggest that automating AI R&D could significantly compress AI progress, outpacing society’s ability to adapt and straining frontier risk mitigations.
Welcome to The Frontier Scenario, an illustrative, fictional account of frontier AI development and how ARI’s Blueprint for Frontier AI Governance could be applied to address AI risks.
The controversial U.S. spy program–FISA Section 702–expired over two months ago. Without a deadline pressuring a clean reauthorization, Congress has a moment to address the loopholes that allow warrantless surveillance of Americans.
